camunda-connectors-development
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents the development of inbound and outbound connectors that ingest and process data from external, untrusted sources such as HTTP webhooks, message brokers (Kafka, RabbitMQ, SQS), and external API polling. This data is eventually interpolated into the process engine's context.
- Ingestion points: External data enters via the
activatemethod inInboundConnectorExecutable(referenced inreferences/connector-sdk-inbound.md) and through@Variableparameter bindings inOutboundConnectorProvider(referenced inreferences/connector-sdk-outbound.md). - Boundary markers: The instructions describe the use of FEEL (Friendly Enough Expression Language) for mapping and transforming data (e.g.,
resultExpression), which can act as a logic boundary but does not inherently prevent the processing of malicious instructions within the data values themselves. - Capability inventory: Connectors built using these instructions have broad capabilities, including performing arbitrary network requests via Java's
HttpClient, modifying process variables, and triggering process instances or message correlations within the Camunda engine. - Sanitization: The skill explicitly recommends and provides examples for using Jakarta Validation annotations (e.g.,
@NotEmpty,@Email,@Pattern) and Hibernate Validator to sanitize and validate inputs before they are processed by the connector logic.
Audit Metadata